Image Analysis for Classifying Forest Fire Using Deep Learning

Abstract

80% of all living things on earth depend on forests as part of their ecosystem for food and shelter. For most of their existence, Homo humans lived in forests. Forests are at risk from forest fires. Because woods exist, there will always be forest fires. Uncontrolled forest fires occur in foliage that is taller than 1.8 meters (6 feet). Researchers’ interest in the topic of fire detection in pictures using computer vision and image processing techniques has significantly increased during the last few years. This study compares various deep learning-based algorithms for detecting forest fires. The dataset is classified using five different DL methods: VGG 16, Inception V3, VGG19, MobileNetV2, and DenseNet201. The dataset, which includes 18,344 images divided into four groups (fire, nonfire, smoke, and fog). Forest fire forecasts are more accurate, according to the experimental investigation, with DenseNet201 having the highest accuracy (96.40%).